While there has long been a general consensus that structural modeling could distinguish observed from true effects, until 1978 procedures for isolating true effects were believed to depend on estimates of reliability or error variance and were typically made from extraexperimental information; now, appropriate latent variable design can effectively isolate the relevant effects of interest. The author uses the example of the evaluative study of the quasi- experimental Head Start program to illustrate these points. Another study is used to illustrate how structural modeling can help analyze intervention data in terms of pretest versus treatment effects. Because nearly all prevention/treatment studies have a serious problem of attrition, structural modeling can help control for missing data, particularly when the missing data contain a few predominantly patterns of missing data. Structural modeling can also be used to identify indicators of program participation in order to assess intervention effectiveness. 2 figures, 14 references, and 1 appendix